Unbiased rheological properties determined by adversarial training with Bingham equation

The Bingham model is generally used to describe the flow of cement-based materials, and its parameters, such as yield stress and plastic viscosity, are measured using a rheometer. However, the rheological measurement does not provide the unique Bingham parameters for a single material when rheometers take different rheological geometries or measuring protocols. This study constructs a model that can yield the ideal Bingham parameters with the rheological measurement. We first introduce the generation of an ideal domain strictly following the Bingham equation, and then an unsupervised domain adaptation by adversarial training makes it possible to match the rheological measurement with the ideal Bingham parameters. The proposed model is applied to the experimental data measured with mortar samples, where the measurements for a single sample are conducted by three different measuring protocols. The resultant (ideal) Bingham parameters are identical regardless of the protocols used.